Palliative Non-Operative Management in Geriatric Hip Fracture Patients: When Would Surgeons Abstain from Surgery?

Abstract

Background: For hip fracture patients with a limited life expectancy, operative and palliative non-operative management (P-NOM) can yield similar quality of life outcomes. However, evidence on when to abstain from surgery is lacking. The aim of this study was to quantify the influence of patient characteristics on surgeons' decisions to recommend P-NOM. Methods: Dutch surgical residents and orthopaedic trauma surgeons were enrolled in a conjoint analysis and structured expert judgement (SEJ). The participants assessed 16 patient cases comprising 10 clinically relevant characteristics. For each case, they recommended either surgery or P-NOM and estimated the 30-day postoperative mortality risk. Treatment recommendations were analysed using Bayesian logistic regression, and perceived risks were pooled with equal and performance-based weights using Cooke's Classical Model. Results: The conjoint analysis and SEJ were completed by 14 and 9 participants, respectively. Participants were more likely to recommend P-NOM to patients with metastatic carcinomas (OR: 4.42, CrI: 2.14-8.95), severe heart failure (OR: 4.05, CrI: 1.89-8.29), end-stage renal failure (OR: 3.54, CrI: 1.76-7.35) and dementia (OR: 3.35, CrI: 1.70-7.06). The patient receiving the most P-NOM recommendations (12/14) had a pooled perceived risk of 30-day mortality between 50.8 and 62.7%. Conclusions: Overall, comorbidities had the strongest influence on participants' decisions to recommend P-NOM. Nevertheless, practice variation and heterogeneity in risk perceptions were substantial. Hence, more decision support for considering P-NOM is needed.

Keywords: conjoint analysis; decision-making; frailty; geriatrics; hip fractures; palliative non-operative management; structured expert judgement.

Figure A1

Trace plots and densities of the posterior distributions of the coefficients (1/3).

Figure A2

Trace plots and densities of the posterior distributions of the coefficients (2/3).

Figure A3

Trace plots and densities of the posterior distributions of the coefficients (3/3).

Figure A4

Autocorrelation plots of the Markov chains for each regression coefficient.

Figure A5

Histograms of the posterior samples drawn through Markov Chain Monte Carlo sampling for each regression coefficient.

Figure 1

Power curves for the attribute levels used in the vignettes. The respective odds ratios (ORs), which were assumed during the power calculations, are listed behind each attribute level.

Figure 2

Overview of participants’ responses to the 14 calibration questions. The dots represent participants’ best estimates of the 30-day mortality rates and the horizontal bars represent their 90% credible intervals. The true 30-day mortality rate is reported in parentheses and depicted by the vertical line. Participants with calibration scores above 0.05 are highlighted in blue.

Figure 3

Distribution of participants’ responses to the 16 vignettes. Vignettes were sorted in descending order of mean 30-day mortality risk. (a) Overview of recommended treatments per vignette, subcategorised by participants’ confidence in the optimality of the elected treatment. (b) Boxplots of the estimated 30-day mortality risks per vignette. Circles denote the mean probabilities (equal weights) and crosses denote the performance-weighted pooled estimates.